Rooftop solar adoption in Washington DC

Author

Kumar H

Published

March 2, 2024

As of 2022, solar energy provided about 3.4% of total U.S. utility-scale electricity (U.S. Energy Information Administration, 2024) Among US cities, Honolulu tops the chart, with per capita solar photovoltaic (PV) capacity installed at 1133 Watts DC/person in 2022. Washington DC was at 10th rank by 2022 (leaping up from 17th in 2018.) (Frontier Group, 2022)

The most likely reason behind this boost was DC’s Renewable Portfolio Standards (RPS) policies. RPS policies are regulatory mandates to increase the production of energy from renewable sources, requiring electric utilities to ensure that a specified percentage of the energy they generate or sell comes from renewable sources by a certain date. Despite the lack of a comprehensive federal RPS, state-level mandates have collectively contributed to a significant increase in the country’s renewable energy capacity, transforming the energy landscape and paving the way for a more sustainable future.

DC’s RPS is one of the country’s most ambitious, as underscored by the Clean Energy DC Omnibus Amendment Act of 2018. This act mandates that 100 percent of the District’s electric supply be sourced from renewable generators by 2032. Additionally, it sets forth a solar “carve-out” requirement that 5.5 percent of the 2032 electric supply must come from in-District solar, with this percentage increasing to 10 percent by 2041. This places the District as the third jurisdiction in the U.S., following Hawai’i and California, to establish a 100 percent renewable energy requirement.

The primary mechanism by which the DC Department of Energy and Environment (DOEE) influences the solar market is through Solar Renewable Energy Credits (SRECs) and Alternative Compliance Payments (ACP).

SRECs and ACP
  • SREC: Market-based instruments that represent the environmental benefits of generating solar power. Each SREC certifies that one megawatt-hour of solar electricity has been generated and can be sold or traded, providing an additional revenue stream for solar energy producers on top of the electricity they sell.

  • ACP: Penalties that electric utilities or other entities must pay if they fail to meet their renewable energy targets, specifically the portion of those targets that must be met through solar energy.

There have been several major policy changes in DC that affect solar adoption incentives, as summarized in this RPS report by the DC Public Service Commission (https://dcpsc.org/Orders-and-Regulations/PSC-Reports-to-the-DC-Council/Renewable-Energy-Portfolio-Standard.aspx)

Installed solar capacity has skyrocketed in DC.

Existing empirical evidence finds that US states that closed off their SREC markets to out-of-state facilites witnessed higher solar installation, as DC did in 2011. (Cohen et al 2022)

However, as the following animated graph shows, adoption really took off around 2016.

Sample gganimate code
# Animate the plot
anim <- p + transition_reveal(date) + 
  enter_grow() +  # Animation for entering points
  exit_fade() +   # Animation for exiting points
  ease_aes('cubic-in-out') + # Animation speed
  labs(title = 'Year: {frame_along}', subtitle = "Tier-One Renewable Percent") # Update title with year

# Render the animation
animate(anim, duration = 10, fps = 30, width = 800, height = 600, renderer =gifski_renderer())

Measuring actual generation, as opposed to capacity installed, is trickier. In 2022, the total retail electricity sales reported by suppliers in the District were 10.16 million MWh. To put this in perspective, Hawaii’s consumption in 2022 was around 9 million MWh, despite being 60 times larger in area and with 2x the population! (EIA 2022)

“The total number of SRECs retired for compliance was 263,919, which is equivalent to nearly 264,000 MWh and represents about 2.6% of the total retail electricity sales for the year”​​. (DC Public Service Commission, 2023) Of course, SRECs can be held before they are cashed - it indicates that by this date, there has been at least 264,000 MWh generation of solar.

The DC Solar Market - Incentives and Costs

In Washington DC, the average solar panel costs $11,000-$14,000 after federal incentives

A household in DC can factor in three types of incentives available for a solar installation :

  • Federal rebates

  • SREC prices

  • Net metering

There is another type of indirect incentive, which might function as a ‘stick’ to the above ‘carrots’. This is the Building Energy Benchmarking Policy (BEPS), which aims to identify and impose financial penalties on large buildings (>10,000 sq feet) with greater than average energy consumption

The following figures describe the outlook for solar generation in the District:

  • Technical potential = 1200 MW
  • Target by 2040 = 665 MW
  • Adoption in 2023 = 201 MW

The above figures are from a report by Synapse Energy Economics, where it also lays out the factors holding back adoption in DC. For now, we zoom in on one particular factor laid out in the report:

“Given that 60 percent of households in the District are renter-occupied, home ownership and upfront costs are likely strong barriers to solar adoption”

Creating a building-level dataset to examine solar adoption

With the help of DC Open Data’s frequently updated and granular data on buildings, we can merge data solar and building attributes data for almost all of DC’s >100,000 residential buildings.

Note: Merging properties with SSL

Linking these two datasets is more complex than a single merge. Solar records are given for an address, but CAMA records are linked to Square Suffix Lot (SSL) numbers, which identifies the plot of land on which a building is located. Some lots have multiple buildings, and some buildings/complexes span multiple lots.

The following table shows a small sample of the rows and columns in our resulting dataset.

SSL STYLE_D ROOF_D NBHDNAME cap_kw max_sun_hours min_dc_kwh
5976 0009 2.5 Story Fin Shingle Congress Heights 0 1605.675 1589.752
3568 0052 2 Story Built Up Eckington 0 1485.294 1450.584
3033 0066 2 Story Metal- Sms Columbia Heights 4.875 1622.292 1619.482
4327 1064 2 Story Shingle Fort Lincoln 3.06 NA NA
3870 0041 2 Story Shingle Brentwood 15.925 1582.055 1579.705

Two variables have been called from the Google Solar API for these building locations:

  • max_sun_hours: Maximum number of sunshine hours received per year, by any point on the roof

  • min_dc_kwh: How much sunlight energy the roof captures in (direct current) DC kWh, assuming a certain minimum number of panels. (This is a transformed variable)

The above example illustrates the fundamental prediction problem facing us - will these buildings adopt?

They seem to be as suitable as the buildings in the bottom 3 rows, receiving high amounts of annual sunlight (from the supply side).

The fact that the solar data has both the exact location and date of solar panel installation, added with DC Open Data’s rich repository of spatial building-level datasets implies that we can drill down at the hyperlocal level for insights in solar adoption patterns.

Modeling paradigms for solar adoption:

Now that the data is consolidated, the next task is to model its diffusion process, and forecast it into the future.

A popular approach is NREL’s Distributed Renewable Generation model (dGen) model.

In a review of around 200 papers, Alipour et al (2021) found the types of modeling approaches:

  • Qualitative: 8.2%

  • Statistical analysis : 56.7%

  • Simulation-based (ABMs, spatial, diffusion): 35.1%

1. Statistical approaches

The variables must account for these factors:

  • Spatial similarity or diffusion frameworks

  • Potential generation (revealed by the Google Solar API)

  • Actual energy use: This is the most difficult to get. Closest we can get is energy expenditures, and that only exists.

Techniques:

  1. OLS and panel regressions
  2. Spatial regression models
  3. Tree-based estimation (XGBoost)
  4. Graph learning

Synapse Inc.(2020) mentions that the high percentage of renters in the district means the incentive for households to spend on solar installation is low.

As a warm-up empirical test, let’s test this out using data from the American Community Survey (ACS) 2020?

Example: Using Census data to check the relation between home-ownership and solar capacity added

Ideally, we would have joined building-level data of owner vs renter occupation, but in its absence, we can use block-group level data from the ACS 2019 on homeownership rates. What does the ACS 2019 show about the characteristics of high-solar adoption areas?

Sample code: tidyCensus

Sample tidyCensus code
st_acs <- get_acs(variables = c(total_pop = "B02001_001E", 
                                  owner_occupied = "B25003_002E",
                                  total_occupied = "B25003_001E"),
                    geography = "block group",
                    geometry = TRUE,
                    state = "DC", 
                    year = 2019) 

The above graph clearly shows that block groups with a higher share of homes inhabited by owners had higher per capita solar capacity, expressed as KiloWatts per 1000 people.

Check out the full empirical modeling here [Coming Soon].

2. Agent based models: NREL dGen

The National Renewable Energy Lab’s Distributed Generation Market Demand (dGen) model is defined as:

“Statistically representative residential and commercial agents are ingested by the model and used to forecast PV adoption based on user specified configurations like electricity rate prices, electricity load growth, solar resource factors, and much more.”

Alipour et al (2021) describe:

self-governing households at the bottom of a social phenomenon where each decides heterogeneously and heuristi cally over time and space, but their aggregated decisions produce macro outcomes

The model is available open-source on Github (linked above) as a deployable docker container on Github.

Do we use the model off the shelf, or do we tweak and adapt it to our case? Stay tuned for the next post! (COMING SOON).